语音识别中基于两层词法树的跨词搜索算法  被引量:1

Cross-word search algorithm based on two-layer lexical tree for speech recognition

在线阅读下载全文

作  者:张国亮[1] 徐明星[1] 李净[1] 郑方[1] 吴文虎[1] 

机构地区:[1]清华大学计算机科学与技术系,北京100084

出  处:《清华大学学报(自然科学版)》2003年第7期981-984,共4页Journal of Tsinghua University(Science and Technology)

基  金:国家自然科学基金资助项目(19871045)

摘  要:为了在连续语音识别过程中充分并且高效地使用上下文相关声学模型,提出了一种新颖的基于两层词法树的跨词搜索算法。采用两层词法树来表示搜索空间,解决了现有单层词法树的规模爆炸问题,使其有能力在词边界搜索中高效地使用上下文相关声学模型进行匹配,充分发挥上下文相关声学模型较好地描述协同发音现象的能力。实验结果表明,与词内搜索算法相比误识率平均下降60%,搜索时间达到实时,证明基于两层词法树的跨词搜索算法具有很好的识别性能。A cr oss-word search algorithm was developed based on a two-layer lexical tree usin g context-dependent acoustic models for efficient continuous speech recognition . The algorithm uses the two-layer lexcial tree to maintain the search network to keep the search space from becoming more complicated during the cross-word s earch. The search algorithm can then use context-dependent models on the word b oundary for good performance. Experimental results show that the search algorith m can work in real-time with 60% less word errors compared with the intra-word search algorithm, which illustrates the efficiency of the cross-word search al gorithm based on the two-layer lexical tree.

关 键 词:语音识别 两层词法树 跨词搜索算法 上下文相关声学模型 语音信号处理 

分 类 号:TN912.34[电子电信—通信与信息系统]

 

参考文献:

正在载入数据...

 

二级参考文献:

正在载入数据...

 

耦合文献:

正在载入数据...

 

引证文献:

正在载入数据...

 

二级引证文献:

正在载入数据...

 

同被引文献:

正在载入数据...

 

相关期刊文献:

正在载入数据...

相关的主题
相关的作者对象
相关的机构对象